用聚类实现无需数据和训练的语音模型压缩，效果优于传统剪枝。
Towards Data-free and Training-free Compression for Speech Foundation Models Using Parameter Clustering
- 通过通道级k-means聚类实现参数压缩，无需微调或新数据。
- 在HuBERT-large上50%稀疏度下，测试集词错率降低超27%。
- 适用于追求轻量化部署的语音模型研究者与工程师。
本文提出一种全新的、无需数据且无需训练的语音基础模型压缩方法，基于通道级k-means聚类实现参数压缩，并探索了按层变化聚类数的更精细混合稀疏性剪枝策略。在LibriSpeech数据集上的实验表明，在HuBERT-large模型上以50%稀疏度进行压缩时，未微调前在test-clean和test-other子集上的词错率（WER）分别比基于幅度剪枝降低了27.73%和18.61%（相对降低34.37%和21.91%），微调3轮后仍保持0.19%和0.79%的绝对降低（相对降低3.36%和4.62%）。在Whisper-large-v3模型上，10%稀疏度下相较幅度剪枝实现2.86%和5.02%的绝对降低（相对降低59.21%和55.29%），且未引入显著的词错率增加，与原始未压缩基线相比性能持平。
原文摘要 · Abstract (English)
This paper presents a novel data-free and training-free compression approach for speech foundation models using channelwise clustering via k-means. More fine-grained, mixed sparsity pruning by layer-level varying number of parameter clusters is also explored. Experiments conducted on the LibriSpeech dataset suggest that when operating with pruning sparsity of 50% on HuBERT-large, consistent WER reductions of 27.73%/18.61% absolute (34.37%/21.91% relative) over the magnitude-based pruning were obtained on the test-clean and test-other subsets before fine-tuning and 0.19%/0.79% absolute (3.36%/4.62% relative) after fine-tuning with only 3 epochs. Similar WER reductions of 2.86%/5.02% absolute (59.21%/55.29% relative) were observed against magnitudebased pruning on Whisper-large-v3 at 10% sparsity, all with no significant WER increase relative to the uncompressed baseline.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。